I’ve been experimenting with generative AI lately – both in my day job and on personal projects – and I thought it was time to jot down some reflections. Not a deep think piece, just a few observations about how tools like Copilot and ChatGPT are starting to shape the way I work.

In my professional life, I’ve used AI to draft meeting agendas, prepare documents, sketch out presentation outlines, and summarise lengthy reports. It’s a co-pilot in the truest sense – it doesn’t replace me, but it often gives me a head start. That said, the results are hit and miss, and I never post anything AI-generated without editing. Sometimes the AI gives me inspiration. Other times, it gives me American spelling and questionable grammar.

But outside work is where things got interesting.

I accidentally vibe coded

It turns out there’s a name for what I’ve been doing in my spare time: vibe coding.

First up, I wanted to connect a microcontroller to an OLED display and to control the display with a web form and a REST API. I didn’t know exactly how to do it, but I had a vague idea. I asked ChatGPT. It gave me code, wiring instructions, and step-by-step guidance to flash the firmware. It didn’t work out of the box – but with a few nudges to fix a compilation error and rework the wiring, I got it going.

Then, I wanted to create a single-page website to showcase a custom GPT I’d built. Again, ChatGPT gave me the starter template. I published it to Azure Static Web Apps, with GitHub for source control and a CI/CD pipeline. All of it AI-assisted.

Both projects were up and running quickly – but finishing them took a lot more effort. You can get 80% of the way with vibes, but the last 20% still needs graft, knowledge, or at the very least, stubborn persistence. And the 80% is the quick part – the 20% takes the time.

What is vibe coding?

In short: it’s when you code without fully knowing what you’re doing. You rely on generative AI tools to generate snippets, help debug errors, or explain unfamiliar concepts. You follow the vibe, not the manual.

And while that might sound irresponsible, it’s increasingly common – especially as generative AI becomes more capable. If you’re solving a one-off problem or building a quick prototype, it can be a great approach.

I should add some context: I do have a Computer Studies degree, and I can code. But aside from batch scripts and a bit of PowerShell, I haven’t written anything professionally since my 1992/93 internship – and that was in COBOL.

So, yes, I have some idea of what’s going on. But I’m still firmly in vibe territory when it comes to ESP32 firmware or HTML/CSS layout.

The good, the bad, and the undocumented

Vibe coding has clear advantages:

• You can build things you wouldn’t otherwise attempt.
• You learn by doing – with AI as your tutor.
• You get to explore new tech without wading through outdated forum posts.

But it also has its pitfalls:

• The AI isn’t always right (and often makes things up).
• Debugging generated code can be a nightmare.
• If you don’t understand what the code does, maintaining it is difficult – if not impossible.
• AI doesn’t always follow best practices – and those change over time.
• It may generate code that’s based on copyrighted sources. Licensing isn’t always clear.

That last pair is increasingly important. Large language models are trained on public code from the Internet – but not everything online is a good example. Some of it is outdated. Some of it is inefficient. Some of it may not be free to use. So unless you know what you’re looking at (and where it came from), you risk building on shaky ground.

Where next?

Generative AI is changing how we create, code, and communicate. But it’s not a magic wand. It’s a powerful assistant – especially for those of us who are happy to get stuck in without always knowing where things will end up.

Whether I’ve saved any time is up for debate. But I’ve definitely done more. Built more. Learned more.

And that feels like progress.

A version of this post was originally published on the Node4 blog.

Featured image by James Osborne from Pixabay.